Cursor turns cloud agents into event-driven workers and brings code hosting into the loop

Cursor’s August 19 cloud-agent release adds subscriptions that can start agents from pull-request activity, Slack messages and scheduled tasks, plus a `/goal` command for persistent task context, custom modes, and subagents that run on their own virtual machines. Two days earlier, Cursor introduced Origin, an early-beta code-hosting product for paid plans with native repositories, pull requests, code browsing and GitHub synchronization. Together with Cursor’s August 13 prebuilt cloud-agent environments, the changes move Cursor from an editor-centered assistant toward a more persistent, event-driven development system.

The important shift is workflow shape rather than another coding benchmark. An agent can now be invoked by events around a repository, keep a longer-lived goal, delegate parallel work and operate close to the code-hosting surface where reviews happen. That reduces the amount of custom orchestration teams need to build for unattended maintenance, review follow-up or scheduled engineering work. It also concentrates more of the software-delivery loop inside Cursor, increasing the importance of permission design, automation-loop safeguards, portability and a clear source-of-truth strategy while Origin remains beta.

Cursor says cloud agents can now subscribe to pull-request events, Slack messages and schedules rather than relying only on a developer manually starting a task. The pull-request path is particularly consequential because an agent can be brought back into work when review activity changes the state of a task. This turns the agent into an event-driven worker embedded in an existing development loop rather than a one-shot coding session.

The same release adds a `/goal` command that gives an agent persistent task context and lets cloud agents use custom modes. Cursor also says subagents now run on their own virtual machines. That separation makes parallel work and specialized delegation more practical, but it also means teams need explicit boundaries around credentials, repository permissions, external services and what happens when multiple agents touch overlapping code.

Cursor introduced Origin on August 17 as an early-beta code-hosting layer available on paid plans. It includes repositories, pull requests and code browsing, can synchronize with GitHub, and is positioned as infrastructure that will support more agent-native workflows. Cursor also lists integrations with Vercel, Depot and Buildkite. For teams that adopt it, the product could shorten the path from an agent noticing an event to creating and reviewing a change, but it also creates a new hosting dependency that should not yet be treated as a drop-in replacement for mature repository governance.

Cursor’s August 13 build-system update adds prebuilt cloud-agent environments so dependencies and setup work can be prepared before an agent starts. Cursor claims roughly three-times faster time to first token and says agents can fall back to the last known-good build when the newest build fails. These are vendor-reported performance improvements rather than independent benchmarks, but the mechanism matters: unattended agents become more useful when environment bootstrapping is predictable and repeatable.

Event-driven agents can create feedback loops if a bot comment, pull-request update or CI action triggers another agent run that causes another event. Builders should treat trigger filters, concurrency, idempotency and human approval as first-class workflow controls. Teams experimenting with Origin should also preserve an export or synchronization path and decide which host is authoritative for protected branches, reviews, CI status and recovery while the product is still early beta.